New formulations for active learning
Author(s)
Ganti Mahapatruni, Ravi Sastry
Advisor(s)
Gray, Alexander G.
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
In this thesis, we provide computationally efficient algorithms with provable statistical guarantees, for the problem of active learning, by using ideas from sequential analysis. We provide a generic algorithmic framework for active learning in the pool setting, and instantiate this framework by using ideas from learning with experts, stochastic optimization, and multi-armed bandits. For the problem of learning convex combination of a given set of hypothesis, we provide a stochastic mirror descent based active learning algorithm in the stream setting.
Sponsor
Date
2014-01-10
Extent
Resource Type
Text
Resource Subtype
Dissertation